Skip to Main content Skip to Navigation

Modélisation des métadonnées multi sources et hétérogènes pour le filtrage négatif et l'interrogation intelligente de grands volumes de données : application à la vidéosurveillance

Franck Jeveme Panta 1
1 IRIT-SIG - Systèmes d’Informations Généralisées
IRIT - Institut de recherche en informatique de Toulouse
Abstract : Due to the massive and progressive deployment of video surveillance systems in major cities, a posteriori analysis of videos coming from these systems is facing many problems, including the following: (i) interoperability, due to the different data (video) formats and camera specifications associated to each system; (ii) time-consuming nature of analysis due to the huge amount of data and metadata generated; and (iii) difficulty to interpret videos which are sometimes incomplete. To address these issues, the need to propose a common format to exchange video surveillance data and metadata, to make video content filtering and querying more efficient, and to facilitate the interpretation of content using external (contextual) information is an unavoidable concern. Therefore, this thesis focuses on heterogeneous and multi-source metadata modeling in order to propose negative filtering and intelligent data querying, which are applicable to video surveillance systems in particular and adaptable to systems dealing with large volumes of data in general. In the applicative context of this thesis, the goal is to provide human CCTV operators with tools that help them to reduce the large volume of video to be processed or viewed and implicitly reduce search time. We therefore initially propose a so-called "negative" filtering method, which enables the elimination from the mass of available videos those that it is know in advance, based on a set of criteria, that the processing will not lead to any result. The criteria used for the proposed negative filtering approach are based on metadata modeling describing video quality and usability/usefulness. Then, we propose a contextual enrichment process based on metadata from the context, enabling intelligent querying of the videos. The proposed contextual enrichment process is supported by a scalable metadata model that integrates contextual information from a variety of sources, and a multi-level query mechanism with a spatio-temporal reasoning ability that is robust to fuzzy queries. Finally, we propose a generic metadata modeling of video surveillance metadata integrating metadata describing the movement and field of view of cameras, metadata from content analysis algorithms, and metadata from contextual information, in order to complete the metadata dictionary of the ISO 22311/IEC 79 standard, which aims to provide a common format to export data extracted from video surveillance systems. The experiments performed using the framework developed in this thesis showed the reliability of our approach in a real case and enabled the validation of our proposals.
Document type :
Complete list of metadata
Contributor : Abes Star :  Contact Connect in order to contact the contributor
Submitted on : Friday, January 22, 2021 - 9:46:07 AM
Last modification on : Tuesday, October 19, 2021 - 2:23:36 PM
Long-term archiving on: : Friday, April 23, 2021 - 7:00:28 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03118294, version 1


Franck Jeveme Panta. Modélisation des métadonnées multi sources et hétérogènes pour le filtrage négatif et l'interrogation intelligente de grands volumes de données : application à la vidéosurveillance. Intelligence artificielle [cs.AI]. Université Paul Sabatier - Toulouse III, 2020. Français. ⟨NNT : 2020TOU30098⟩. ⟨tel-03118294⟩



Record views


Files downloads